A method for intelligent monitoring of operation status of intelligent service terminal
By collecting and combining the historical dimension operation data of the intelligent service terminal, using the bounding factors and correlation conditions to monitor abnormal data, the clustering error problem caused by changes in the operating status of the intelligent service terminal is solved, and detection efficiency and accuracy are improved.
Patent Information
- Application Number
- CN202410930952.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-12
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-07-12
AI Technical Summary
Changes in the operating state of the intelligent service terminal lead to changes in the association relationship of operation data in different dimensions in different times. The connected graph obtained by the traditional CABDDCG based on the dynamic splitting of the connectivity graph has error information, which reduces the accuracy and detection efficiency of the clustering results. .
By collecting historical dimension operation data, obtaining boundary factors based on the changes in the operation cycle and trend differences, dividing the operation data segments, and combining them with the correlation of data from different dimensions, using node structures to monitor abnormal data to improve the accuracy of clustering results.
By adaptively combining and monitoring the running data in historical dimensions, the possibility of wrong nodes of traditional clustering algorithms is reduced, and the detection efficiency of running data of intelligent service terminals and the accuracy of clustering results is improved.
Smart Images

Figure CN118467301B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data processing technology, and more specifically to a method for intelligent monitoring of the operation status of smart service terminals. Background Technology
[0002] Existing smart service terminals primarily improve service quality by analyzing multi-dimensional data to understand user needs. Therefore, to ensure the service quality of smart service terminals, it is necessary to monitor their multi-dimensional data.
[0003] Traditional methods typically utilize CABDDCG (Clustering Algorithm Based on Dynamic Division of Connected Graph) clustering algorithm, which uses dynamic division of connected graphs, for intelligent monitoring of multi-dimensional data. However, the amount of data transmitted by intelligent service terminals changes continuously with users' personalized needs, causing the operational status of the intelligent service terminals to change constantly. This results in different correlations between operational data of different dimensions at different times, with some abnormal nodes being close to normal nodes. Consequently, the connected graph obtained by the CABDDCG clustering algorithm contains some erroneous information, reducing the accuracy of the clustering results and thus reducing the efficiency of detecting the operational data of intelligent service terminals. Summary of the Invention
[0004] This invention provides an intelligent monitoring method for the operation of smart service terminals to solve the existing problem: because the operating status of smart service terminals is constantly changing, there are different correlations between different dimensions of operating data at different times, resulting in some erroneous information in the connected graph obtained by the CABDDCG clustering algorithm based on dynamic splitting of connected graphs.
[0005] The intelligent monitoring method for the operation of intelligent service terminals of the present invention adopts the following technical solution:
[0006] Includes the following steps:
[0007] Collect historical data for each data category over several days, with each historical data point corresponding to a record time.
[0008] Based on the changes in the running cycle and the differences in the running trend of the historical dimension running data of the same data type on the same day, the boundary factor of each historical dimension running data is obtained; based on the boundary factor, the running data of each data type of the same dimension in several days is divided into several running data segments.
[0009] Based on the correlation between different data types of the same sequence number running data segments, the changes in data running patterns, the boundary factors of each historical dimension running data and the recording time, several data types of dimensions are combined to obtain several combinations of data types of each dimension.
[0010] Based on the similarity of the connection edges between the node structures corresponding to each historical dimension's running data and other historical dimension's running data in the node structure of several historical dimension's running data combinations, several abnormal dimension's running data are obtained by monitoring several historical dimension's running data.
[0011] Preferably, the method for obtaining the boundary factor for each historical dimension's operational data based on the operational cycle changes and operational trend differences of the same dimension's operational data within the same day includes:
[0012] Preset an initial window size One step and a termination window size Set the initial window size. As the starting point for window size, the step size is... Gradually increase the window size until the increased window size meets the requirements. Stop expanding the window at any time, obtain the window size during each expansion process, and record the window size during each expansion process as the time decomposition window size;
[0013] For any given time decomposition window size, obtain the next... The data categories of each dimension are in the first Several clusters under each decomposition item; obtain the first cluster under the time decomposition window size. The data categories of each dimension are in the first Similar clusters of each cluster under each decomposition item;
[0014] In the The data categories of each dimension are in the first Tianzhongdi In the cluster to which the decomposition term belongs, except for the first one... The data categories of each dimension are in the first Tianzhongdi The set consisting of all the decomposition terms other than the first decomposition term is denoted as the i-th decomposition term. The data categories of each dimension are in the first Tianzhongdi The reference decomposition set of the nth decomposition item; according to the nth decomposition item The data categories of each dimension are in the first The historical dimensions of the data contained in different clusters under different decomposition items of a given day, the reference decomposition set, and the corresponding similar clusters are used to obtain the data under the time decomposition window size. The data categories of each dimension are in the first Decomposition window weights for each day;
[0015] For the The data categories of each dimension are in the first The first in the sky The historical dimension of operational data, based on the first The data categories of each dimension are in the first The numerical changes of all historical dimensions of the data in Tianzhong, and the values of each time decomposition window size. The data categories of each dimension are in the first The decomposition window weights for each day are obtained. The data categories of each dimension are in the first The first in the sky The degree of boundary of the historical dimension running data; obtain the degree of boundary of the historical dimension running data of all dimension data types in each day, linearly normalize all the degree of boundary, and record each normalized degree of boundary as the boundary factor.
[0016] Preferably, the acquisition time decomposition window size is the first... The data categories of each dimension are in the first Several clusters under each decomposition term, including the following specific methods:
[0017] Using the time decomposition window size as the window size, and based on the window size, the STL time series decomposition algorithm is used to decompose the th... The data categories of each dimension are in the first Decompose the sequence of all historical dimensions of data in Tianzhong to obtain the first... The first day The data types in each dimension include trend terms, periodic terms, and residual terms; and the first... The data categories of each dimension are in the first The trend and cycle terms in the sky are denoted as the following numbers respectively: The data categories of each dimension are in the first The decomposition terms in the sky;
[0018] Using the least squares method, for the th The data categories of each dimension are in the first Tianzhongdi Curve fitting was performed on all historical dimensions of the decomposition terms to obtain the first decomposition term. The data categories of each dimension are in the first Tianzhongdi The fitted curve of the nth decomposition term; obtain the nth... The data categories in this dimension rank first among all days. The fitted curve of the i-th decomposition term will be the i-th The data categories of each dimension in any two days The inverse proportional value of the Friesian distance between the fitted curves of the nth decomposition terms is denoted as the decomposition label distance. Using the decomposition label distance as a distance metric, the DBSCAN clustering algorithm is applied to the nth decomposition term. The data categories in this dimension rank first among all days. Clustering the decomposition terms yields the th decomposition term. The data categories of each dimension are in the first Several clusters under each decomposition term.
[0019] Preferably, the acquisition time decomposition window size is the first... The data categories of each dimension are in the first The specific methods for identifying similar clusters for each cluster under each decomposition item are as follows:
[0020] The first The data categories of each dimension are in the first Any cluster under each decomposition term is denoted as the labeled cluster, and each cluster other than the labeled cluster is denoted as the reference cluster. For any reference cluster, the cluster combination formed by the labeled cluster and the reference cluster is denoted as the first cluster combination of the labeled cluster, and the mean of the Friesian distance of the fitted curves between any two decomposition terms between the labeled cluster and the reference cluster is denoted as the similarity coefficient of the first cluster combination.
[0021] Obtain the similarity coefficients of all first cluster combinations, and denote the first cluster combination with the smallest similarity coefficient as the similar first cluster combination; denote the reference cluster in the similar first cluster combination as the similar cluster of the target cluster.
[0022] Preferably, the method according to the first The data categories of each dimension are in the first The historical dimensions of the data contained in different clusters under different decomposition items of a given day, the reference decomposition set, and the corresponding similar clusters are used to obtain the data under the time decomposition window size. The data categories of each dimension are in the first The decomposition window weights for each day include the following specific methods:
[0023] The first The data categories of each dimension are in the first Tianzhongdi All decomposition terms within the reference decomposition set of the i-th decomposition term, and the i-th decomposition term. The data categories of each dimension are in the first Tianzhongdi The mean of the distances between the decomposition identifiers of the nth decomposition items is denoted as the first mean; the nth... The data categories of each dimension are in the first Tianzhongdi All decomposition terms within similar clusters of the cluster to which the i-th decomposition term belongs, and the i-th decomposition term... The data categories of each dimension are in the first Tianzhongdi The average distance between the decomposition markers of the first and second decomposition items is denoted as the second mean; the ratio of the first mean to the second mean is denoted as the third mean. The data categories of each dimension are in the first Tianzhongdi The first ratio of the factorization terms; the first ratio of the factorization terms; The data categories of each dimension are in the first The mean of the first ratios of all decomposition terms in the time decomposition window is denoted as the first ratio under the time decomposition window size. The data categories of each dimension are in the first Decomposition window weights for each day.
[0024] Preferably, the method according to the first The data categories of each dimension are in the first The numerical changes of all historical dimensions of the data in Tianzhong, and the values of each time decomposition window size. The data categories of each dimension are in the first The decomposition window weights for each day are obtained. The data categories of each dimension are in the first The first in the sky The degree of demarcation of historical dimension operating data includes the following specific methods:
[0025] For the The data categories of each dimension are in the first For any two adjacent historical dimension running data points in the sky, the ratio of the difference between the second and first historical dimension running data points to 2 is recorded as the slope of the second historical dimension running data point; obtain the... The data categories of each dimension are in the first The slope of all historical dimensions of the data in Tianzhong;
[0026] Using the autocorrelation function to obtain the time decomposition window size, the first... The period of each data category in each dimension within each day, will be the period of the periodic item. The absolute value of the difference in the periods between any two-day period items for each dimension of data is used as a distance metric. Based on this distance metric, the DBSCAN clustering algorithm is used to cluster the data within the time decomposition window size. Each dimension of data is clustered into a daily periodic item to obtain several clusters, and each cluster is recorded as a dimension periodic cluster.
[0027] The first The first time decomposition window size The data categories of each dimension are in the first Tianzhong, No. The degree of demarcation of the historical dimension's operational data is denoted as the first demarcation degree; the degree of demarcation of the first historical dimension's operational data is denoted as the second demarcation degree. The first time decomposition window size The data categories of each dimension are in the first Tianzhongdi Within the dimensional periodic cluster of the historical dimension's operational data, except for the 1st... Besides the historical dimension of operational data, the first The degree of boundary between the first and second historical dimensions of the data is denoted as the second boundary degree; the absolute value of the difference between the first and second boundary degrees is denoted as the third boundary degree. The first absolute value of the historical dimension's running data; the first... The first time decomposition window size The data categories of each dimension are in the first Tianzhongdi Within the dimensional periodic cluster of the historical dimension's operational data, except for the 1st... The mean of the first absolute value of all historical data (excluding the first historical dimension) is denoted as the third mean; The first time decomposition window size The maximum value of the boundary degree of all historical dimension running data in all the periodic clusters of each dimension data type is recorded as the first maximum value; the ratio of the third mean to the first maximum value is recorded as the second ratio.
[0028] The first The first time decomposition window size The data categories of each dimension are in the first The sum of the residual term of day 1 and 1 is denoted as the first sum; the sum of the residual term of day 2 is denoted as the second sum. The first time decomposition window size The data categories of each dimension are in the first The decomposition window weight for day 1 is denoted as the first weight; the weight for day 2 is denoted as the second weight. The data categories of each dimension are in the first The product of the slope variance, the first weight, and the first sum of all historical dimensions of the data in Tianzhong is denoted as the [missing value]. The first time decomposition window size The data categories of each dimension are in the first The first in the sky The first product of the historical dimensions of the running data; decompose all time into window sizes. The data categories of each dimension are in the first The first in the sky The mean of the first product of the historical data across each dimension is denoted as the _th _ The data categories of each dimension are in the first The first in the sky The degree of demarcation of historical dimension running data.
[0029] Preferably, the specific method for dividing the historical data of each dimension type over several days into several data segments based on the boundary factor includes:
[0030] Preset a threshold factor ; Set the boundary factor greater than The historical data of each dimension is recorded as the initial segment point; the recording time corresponding to each initial segment point is recorded as the segment time point; and the data segment consisting of the historical data of each dimension between adjacent segment time points in each day is recorded as the running data segment of each dimension in each day.
[0031] Preferably, the method for combining several data categories based on the correlation between different data categories with the same sequence number, the changes in data operation patterns, the boundary factors of each historical data category, and the recording time, to obtain several combinations of data categories for each data category, includes the following specific methods:
[0032] For the The data categories of each dimension are in the first The first in the sky The first running data segment will be the first... The absolute value of the difference between the recording time of the first historical dimension running data and the last historical dimension running data in each running data segment is denoted as the nth segment. The runtime of the first data segment; The running time of the first data segment and the first The ratio of all historical dimensions of running data in the first running data segment is denoted as the i-th segment. The single-segment runtime value of the first running data segment; the first segment's runtime value; The data categories of each dimension are in the first The variance of the single-segment runtime value of all running data segments in the sky, denoted as the i-th segment. The data categories of each dimension are in the first The daily running time value in the sky; all dimensional data types in the first The average daily running time of the day is denoted as the th day. Daily data execution level;
[0033] The first The data categories of each dimension are in the first Tianzhongdi The mean of the boundary factors of all historical dimensions of running data in each running data segment is denoted as the first running boundary mean. The mean of the boundary factors of the first running data segment is then used. The data categories of each dimension are in the first Tianzhongdi The mean of the boundary factors of all historical dimension running data in each running data segment is denoted as the second running boundary mean. The mean of the first running boundary mean and the mean of the second running boundary mean are denoted as the third running boundary mean. Day The first of the data categories in the dimension The first running data segment and the first The first of the data categories in the dimension The dimension boundary reference value of the first running data segment; the first Day The data types of each dimension and the first The dimension boundary reference values of all running data segments with the same sequence number between each dimension data type are linearly normalized, and each normalized dimension boundary reference value is recorded as a dimension boundary reference factor.
[0034] Based on the daily data performance, the first [day] of each [day]... The first of the data categories in the dimension The first running data segment and the first The first of the data categories in the dimension The dimension boundary reference factor for the first running data segment, and the first... The data types of each dimension and the first The correlation between data categories of each dimension and the same sequence number segment among different data categories of each dimension on each day is obtained. The data types of each dimension and the first Dimensional correlation factors for each data category;
[0035] Any data category in any dimension is designated as the target data category, and each data category other than the target data category is designated as a reference data category. For any reference data category, a preset threshold for a dimension correlation factor is defined. If the dimensional correlation factor between the target dimension data type and the reference dimension data type is greater than 1, The combination of the target dimension data type and the reference dimension data type is denoted as the dimension type combination of the target dimension data type.
[0036] Preferably, the step of analyzing the daily data execution progress for each day, and then analyzing the data for the first day of each day... The first of the data categories in the dimension The first running data segment and the first The first of the data categories in the dimension The dimension boundary reference factor for the first running data segment, and the first... The data types of each dimension and the first The correlation between data categories of each dimension and the same sequence number segment among different data categories of each dimension on each day is obtained. The data types of each dimension and the first The dimensional correlation factors for each data category include the following specific methods:
[0037] The first Day The first of the data categories in the dimension The first running data segment, and the first The first of the data categories in the dimension The dimension boundary reference factor of the first running data segment is denoted as the first reference factor; the dimension boundary reference factor of the second data segment is denoted as the first reference factor. Day The first of the data categories in the dimension The first running data segment, and the first The first of the data categories in the dimension The Pearson correlation coefficient of the nth running data segment, denoted as the nth... Day The data types of each dimension and the first The first correlation coefficient for each dimension of data; for all days... The data types of each dimension and the first The mean of the first correlation coefficient for each dimension of data is denoted as the fourth mean; the mean of the first correlation coefficient for each dimension of data is denoted as the fourth mean. The product of the inversely proportional normalized value of the daily data performance and the fourth mean is denoted as the [missing value]. Day The data types of each dimension and the first The second product of the data categories in each dimension; [and] the data from all days. The data types of each dimension and the first The mean of the second product of the data categories in each dimension, denoted as the _th The data types of each dimension and the first The degree of dimensional correlation among data categories in each dimension;
[0038] Obtain the degree of dimensional correlation between any two data categories, linearly normalize all dimensional correlations, and record the normalized degree of dimensional correlation as the dimensional correlation factor.
[0039] Preferably, the method for monitoring several historical dimension running data based on the similarity of the connection edges between the node structures corresponding to each historical dimension running data and other historical dimension running data within the node structure, to obtain several abnormal dimension running data, includes the following specific methods:
[0040] Let any one of the data categories be designated as the first target data category. For any combination of data categories within the first target data category, use the historical data of each reference data category within the combination as the x-coordinate of the node, and use the historical data of each historical data category as the y-coordinate of the node. Construct several nodes based on their x-coordinates and y-coordinates. For any two nodes, denote the inversely proportional normalized value of the Euclidean distance between them as the node connection determination distance. A preset threshold for the node connection determination distance is defined. If the node connection distance between these two nodes is greater than... Connect these two nodes and denote the connection as a link; if the node connection distance between the two nodes is less than or equal to... The connection between these two nodes is recorded as a disconnection; the connection between any two nodes is obtained, and the node structure formed by the connection between any two nodes is recorded as a combination node structure of dimension type combination;
[0041] In the first target dimension data category, any historical dimension running data is denoted as target running data, and each historical dimension running data other than target running data is denoted as reference running data. For any reference running data and any dimension category combination of the first target dimension data category, in the combined node structure of the dimension category combination, the node corresponding to the target running data is denoted as the target node, the node corresponding to the reference running data is denoted as the reference node, each node connected to the target node is denoted as the first reference node, each node connected to the reference node is denoted as the second reference node, and each identical node between the first and second reference nodes is denoted as the same-edge node between the target running data and the reference running data. For any same-edge node, the vector formed by the target node and the same-edge node is denoted as the first vector, the vector formed by the reference node and the same-edge node is denoted as the second vector, and the cosine similarity between the first and second vectors is denoted as the first similarity between the target node and the reference node under the same-edge node. The mean of the first similarity between the target node and the reference node under all same-edge nodes is denoted as the node edge similarity between the target running data and the reference running data.
[0042] The first In the combined node structure of each dimension type combination, the number of all nodes with the same edge between the target running data and the reference running data is denoted as the first quantity; the number of nodes with the same edge between the target running data and the reference running data is denoted as the second quantity. The number of connections in the combined node structure of each dimension type is denoted as the second quantity; the ratio of the first quantity to the second quantity is denoted as the third ratio; and the third... In the combined node structure of the three dimensions, the node similarity between the target running data and the reference running data is denoted as the reference similarity; the product of the third ratio and the reference similarity is denoted as the first target dimension data category to which the target running data belongs. The third product of all combinations of dimensions; the mean of the third product of all combinations of dimensions of the first target dimension data to which the target running data belongs is recorded as the nearest neighbor degree between the target running data and the reference running data; the nearest neighbor degree between the target running data and all reference running data is obtained, all nearest neighbor degrees are linearly normalized, and each normalized nearest neighbor degree is recorded as the nearest neighbor factor.
[0043] Preset a nearest neighbor factor threshold The nearest neighbor factor of the target running data is greater than The reference running data is denoted as the nearest running data. All nearest running data of the target running data are obtained, and all nearest running data of the running data of each historical dimension are obtained.
[0044] Based on the Euclidean distance between each nearest neighbor of the historical data and each node in the historical data, the CABDDCG clustering algorithm based on dynamic splitting of the connected graph is used to cluster all historical data, resulting in several final clusters; a preset threshold for the proportion of nodes is used. The number of historical dimension running data is recorded as ,Will and The product of these is denoted as the quantity threshold. For any final cluster, if the number of historical data points in all dimensions within the final cluster is less than... Each historical dimension running data in the final cluster is recorded as an abnormal dimension running data.
[0045] The beneficial effects of the technical solution of this invention are as follows: By combining the changes in the operating cycle of historical dimension operating data within the same dimension data type, and the correlation between historical dimension operating data of different dimension data types, the dimension data types are adaptively combined to obtain several dimension type combinations, thereby monitoring historical dimension operating data; First, based on the changes in the operating cycle and the differences in the operating trend of historical dimension operating data of the same dimension data type on the same day, a boundary factor for each historical dimension operating data is obtained, which is used to reflect the degree of disorder in the data change patterns around the historical dimension operating data, and can better group historical dimension operating data within the same data change pattern into a segment; Then, based on the correlation between operating data segments with the same sequence number and the changes in data operating patterns between different dimension data types, several dimension data types are combined to obtain each dimension data type Several dimensional combinations are used to reflect the correlation between two different data categories, making it easier to determine the relationship between them. Finally, based on the similarity of the connection edges between each historical dimension's running data and the corresponding node structures of other historical dimension's running data within the node structure, the historical dimension's running data is monitored. This invention analyzes the changes in historical dimension's running data within the same data category and the correlation between different data categories, adaptively obtaining reasonable nodes corresponding to each historical dimension's running data, and monitoring the historical dimension's running data based on these nodes. This reduces the possibility of erroneous nodes when the traditional CABDDCG clustering algorithm based on dynamic splitting of connected graphs constructs a connected graph, making the corresponding clustering results more accurate and improving the detection efficiency of the running data of intelligent service terminals. Attached Figure Description
[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0047] Figure 1 This is a flowchart of the steps of the intelligent monitoring method for the operation of an intelligent service terminal according to the present invention;
[0048] Figure 2 This is a flowchart illustrating the characteristic relationships of the intelligent monitoring method for the operation of intelligent service terminals according to the present invention. Detailed Implementation
[0049] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of the intelligent monitoring method for the operation of intelligent service terminals proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.
[0050] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0051] The specific scheme of the intelligent monitoring method for the operation status of intelligent service terminals provided by the present invention will be described in detail below with reference to the accompanying drawings.
[0052] Please see Figure 1 The diagram illustrates a flowchart of a method for intelligent monitoring of the operation status of a smart service terminal according to an embodiment of the present invention. The method includes the following steps:
[0053] Step S001: Collect historical data for several dimensions in each data category.
[0054] It should be noted that traditional methods typically utilize CABDDCG's clustering algorithm based on dynamic splitting of connected graphs for intelligent monitoring of multi-dimensional data. However, the amount of data transmitted by intelligent service terminals changes continuously with users' personalized needs, causing the operational status of intelligent service terminals to change constantly. This results in different correlations between operational data of different dimensions at different times, and some abnormal nodes are close to normal nodes. Consequently, the connected graph obtained by CABDDCG's clustering algorithm based on dynamic splitting of connected graphs contains some erroneous information, reducing the accuracy of clustering results and thus reducing the efficiency of detecting operational data of intelligent service terminals.
[0055] Specifically, the first step is to collect historical operational data. The process involves retrieving historical data from the intelligent service terminal's historical database, covering four data categories: CPU temperature, memory space, network bandwidth, and fan speed. Taking any one category as an example, all historical data within that category are linearly normalized. Each normalized historical data point is then recorded as historical operational data. All historical operational data from all categories are then retrieved. Each category contains multiple historical operational data points, each corresponding to a recording time, and the number of historical operational data points is consistent across all categories. Furthermore, in this embodiment, all historical operational data points within each category are sorted by default in ascending order of recording time. It should be noted that this embodiment does not specifically limit the recording time range, the number of categories, or the types of historical operational data. These limitations can be determined based on the specific implementation.
[0056] Thus, the historical operational data for all dimensions across all data types has been obtained using the methods described above.
[0057] Step S002: Based on the changes in the running cycle and the differences in the running trend of the historical dimension running data of the same dimension type on the same day, obtain the boundary factor of the running data of each historical dimension; based on the boundary factor, divide the running data of each dimension type of historical dimension over several days to obtain several running data segments.
[0058] It should be noted that traditional CABDDCG clustering algorithms based on dynamic splitting of connected graphs convert existing historical dimensional operational data into nodes, construct a connected graph by calculating distance metrics between nodes, and then perform classification monitoring based on the connected graph. However, in reality, user needs are constantly changing, and different nodes within each dimension exhibit different numerical trends and patterns. Furthermore, different dimensions can influence each other to some extent the changing patterns of historical dimensional operational data within each dimension. To improve the intelligent monitoring efficiency of the intelligent service terminal, this embodiment integrates the changing characteristics of historical dimensional operational data across multiple dimensions, segments the historical dimensional operational data within each dimension, and obtains several operational data segments for subsequent analysis and processing.
[0059] Specifically, a preset initial window size is used. One step and a termination window size In this embodiment, This example is used for illustration; no specific limitations are set in this embodiment. This can be determined based on the specific implementation; the initial window size... As the starting point for window size, the step size is... Gradually increase the window size until the increased window size meets the requirements. Stop expanding the window at any time, obtain the window size during each expansion process, and record the window size during each expansion process as the time decomposition window size.
[0060] Furthermore, taking any given time decomposition window size as an example, this time decomposition window size is used as the window size. Based on the window size, the STL time series decomposition algorithm is used to decompose the data for the first... The data categories of each dimension are in the first Decompose the sequence of all historical dimensions of data in Tianzhong to obtain the first... The first day The data types in each dimension include trend terms, periodic terms, and residual terms; and the first... The data categories of each dimension are in the first The trend and cycle terms in the sky are denoted as the following numbers respectively: The data categories of each dimension are in the first The data is divided into two decomposition items for each day. Each dimension of data has two decomposition items for each day. In addition, the process of decomposing the data sequence according to the window size is a well-known part of the STL (Seasonal-Trend decomposition procedure based on Loess) time series decomposition algorithm, and will not be described in detail in this embodiment.
[0061] Furthermore, using the least squares method, for the th The data categories of each dimension are in the first Tianzhongdi Curve fitting was performed on all historical dimensions of the decomposition terms to obtain the first decomposition term. The data categories of each dimension are in the first Tianzhongdi The fitted curve of the nth decomposition term; obtain the nth... The data categories in this dimension rank first among all days. The fitted curve of the i-th decomposition term will be the i-th The data categories of each dimension in any two days The inverse proportional value of the Friesian distance between the fitted curves of the nth decomposition terms is denoted as the decomposition label distance. Using the decomposition label distance as a distance metric, the DBSCAN clustering algorithm is applied to the nth decomposition term. The data categories in this dimension rank first among all days. Clustering the decomposition terms yields the th decomposition term. The data categories of each dimension are in the first Several clusters under each decomposition term. Each cluster contains the first... The data categories of each dimension in a given number of days The decomposition terms are as follows: the least squares method and the acquisition of the Frescher distance are well-known techniques. The specific process of clustering based on the distance metric is the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) clustering algorithm, which is well-known and will not be described in this embodiment.
[0062] Furthermore, the first The data categories of each dimension are in the first Any cluster under the first decomposition term is denoted as the labeled cluster, and each cluster other than the labeled cluster is denoted as the reference cluster. Taking any reference cluster as an example, the cluster combination formed by the labeled cluster and the reference cluster is denoted as the first cluster combination of the labeled cluster. The mean of the Fréchet distances of the fitted curves between any two decomposition terms between the labeled cluster and the reference cluster is denoted as the similarity coefficient of the first cluster combination. The similarity coefficients of all first cluster combinations are obtained, and the first cluster combination with the smallest similarity coefficient is denoted as the similar first cluster combination. The reference clusters in the similar first cluster combination are denoted as similar clusters of the target cluster in the similar first cluster combination. The first cluster combination is then used to obtain the first cluster combination. The data categories of each dimension are in the first Each cluster under each decomposition term is a similar cluster. Each cluster corresponds to one similar cluster.
[0063] Furthermore, in the The data categories of each dimension are in the first Tianzhongdi In the cluster to which the decomposition term belongs, except for the first one... The data categories of each dimension are in the first Tianzhongdi The set consisting of all the decomposition terms other than the first decomposition term is denoted as the i-th decomposition term. The data categories of each dimension are in the first Tianzhongdi The reference decomposition set of the nth decomposition item. The data categories of each dimension are in the first The historical dimensions of the data contained in different clusters under different decomposition items of a given day, the reference decomposition set, and the corresponding similar clusters are used to obtain the data for the first time decomposition under that time decomposition window size. The data categories of each dimension are in the first The decomposition window weights for each day. Each decomposition item corresponds to a reference decomposition set. As an example, the weights of the first day under this time decomposition window size can be calculated using the following formula. The data categories of each dimension are in the first Decomposition window weights for days:
[0064] ;
[0065] In the formula, This indicates the size of the time decomposition window. The data categories of each dimension are in the first Decomposition window weights for each day; Indicates the first The data categories of each dimension are in the first The total number of all decomposition items for a day; Indicates the first The data categories of each dimension are in the first Tianzhongdi The number of decomposition items within the reference decomposition set of each decomposition item; Indicates the first The data categories of each dimension are in the first Tianzhongdi The reference decomposition set of the nth decomposition item The decomposition term, and the first decomposition term The data categories of each dimension are in the first Tianzhongdi The distance between decomposition identifiers of each decomposition item; Indicates the first The data categories of each dimension are in the first Tianzhongdi The number of all decomposition items within similar clusters of the cluster to which each decomposition item belongs; Indicates the first The data categories of each dimension are in the first Tianzhongdi Within the similar clusters of the cluster to which the decomposition item belongs, the i-th The decomposition term, and the first decomposition term The data categories of each dimension are in the first Tianzhongdi The distance between decomposition identifiers of each decomposition item; Indicates the first The data categories of each dimension are in the first Tianzhongdi Intra-class similarity of the clusters to which each decomposition item belongs; Indicates the first The data categories of each dimension are in the first Tianzhongdi The inter-class similarity of the clusters to which each decomposition item belongs. Where the time decomposition window size is less than the... The data categories of each dimension are in the first The larger the decomposition window weight for the first day, the better for the second day. The first dimension of data types In general, the more likely historical data analyzed using this time decomposition window size is to exhibit regularity, the more suitable this time decomposition window size is for analyzing the first dimension. The data categories of each dimension are in the first Historical data for all days. Retrieve all time-decomposition window sizes. The data categories of each dimension are in the first Decomposition window weights for each day.
[0066] Furthermore, with the first The data categories of each dimension are in the first Taking any two adjacent historical dimension running data as an example, in these two historical dimension running data, the ratio of the difference between the second historical dimension running data and the first historical dimension running data to 2 is recorded as the slope of the second historical dimension running data; obtain the first... The data categories of each dimension are in the first The slope of all historical dimensions of the data in the sky. It should be noted that this embodiment defaults to the slope of the first... The data categories of each dimension are in the first The slope of the first historical dimension of the data is 1.
[0067] Furthermore, taking any given time decomposition window size as an example, the autocorrelation function is used to obtain the th time decomposition window size. The period of each dimension of data category in each day's periodic item will be the first... The absolute value of the difference in the periods between any two-day period items for each dimension of data is used as a distance metric. Based on this distance metric, the DBSCAN clustering algorithm is used to perform clustering under this time decomposition window size. Each data category in each dimension is clustered into several periodic clusters for each day, and each cluster is denoted as a dimensional periodic cluster. Each data category in a dimension corresponds to multiple dimensional periodic clusters. Furthermore, the process of obtaining the period of each periodic item is a well-known concept in autocorrelation functions.
[0068] Furthermore, with the first The data categories of each dimension are in the first The first in the sky Taking historical dimension data as an example, according to the first... The data categories of each dimension are in the first The slope of all historical dimensions of the data in Tianzhong, under each time decomposition window size. Dimensional periodic clusters for each data type, and the first cluster under each time decomposition window size. The data categories of each dimension are in the first The decomposition window weights for each day are obtained. The data categories of each dimension are in the first The first in the sky The degree of boundary of the historical dimension of the data. As an example, the boundary can be calculated using the following formula. The data categories of each dimension are in the first The first in the sky The degree of boundary of historical dimension running data:
[0069] ;
[0070] In the formula, Indicates the first The data categories of each dimension are in the first The first in the sky The degree of demarcation of historical dimension operating data; This represents the number of time decomposition window sizes; Indicates the first The first time decomposition window size The data categories of each dimension are in the first Decomposition window weights for each day; Indicates the first The first time decomposition window size The data categories of each dimension are in the first The residual term of the day; Indicates the first The data categories of each dimension are in the first The variance of the slope of all historical dimensions of the data in Tianzhong; Indicates the first The first time decomposition window size The data categories of each dimension are in the first Tianzhongdi Within the dimensional periodic cluster of the historical dimension's operational data, except for the 1st... The number of historical data points other than the historical data points themselves; Indicates the first The first time decomposition window size The data categories of each dimension are in the first Tianzhong, No. The degree of demarcation of historical dimension operating data; Indicates the first The first time decomposition window size The data categories of each dimension are in the first Tianzhongdi Within the dimensional periodic cluster of the historical dimension's operational data, except for the 1st... Besides the historical dimension of operational data, the first The degree of demarcation of historical dimension operating data; Indicates the first The first time decomposition window size The maximum value of the boundary degree of all historical dimension running data in all the periodic clusters of each dimension data type; This indicates taking the absolute value. Where the first... The data categories of each dimension are in the first The first in the sky The greater the degree of boundary between the historical dimensions of the data, the more it indicates that the first historical dimension is more distinct. The more chaotic the patterns of data change around the historical dimension's operating data, the more it reflects the... The more a historical dimension's operational data falls within a region where the pattern of change has undergone significant fluctuations, the more effective the method becomes. For each data category across all dimensions, the method obtains the boundary degree of each historical dimension's operational data for each day. All boundary degrees are then linearly normalized, and each normalized boundary degree is recorded as a boundary factor.
[0071] Furthermore, a threshold for a dividing factor is preset. In this embodiment, This example is used for illustration; no specific limitations are set in this embodiment. This can be determined based on the specific implementation situation; the demarcation factor should be greater than... The historical data for each dimension is recorded as the initial segment point; the recording time corresponding to each initial segment point is recorded as the segment time point; and the data segment consisting of the historical data of each dimension between adjacent segment time points on each day is recorded as the running data segment for each dimension on each day. Each dimension has multiple running data segments on each day.
[0072] At this point, all runtime data segments have been obtained using the methods described above.
[0073] Step S003: Based on the correlation between different data types of the same sequence number running data segments, the changes in data running patterns, the boundary factors of each historical dimension running data and the recording time, combine several data types of dimensions to obtain several combinations of data types of each dimension.
[0074] It should be noted that, in reality, the relationships between nodes across different dimensions are constantly changing. Some nodes may be noisy or contain abnormal data, thus affecting the monitoring efficiency of the intelligent service terminal's operational data. Furthermore, since the historical operational data for each dimension exhibits its own data variation patterns, this embodiment analyzes the dimensional relationships between historical operational data across different dimensions to obtain several operational dimension combinations for subsequent analysis and processing, in order to improve the intelligent monitoring efficiency of the intelligent service terminal.
[0075] Specifically, taking the first The data categories of each dimension are in the first The first in the sky Taking the first running data segment as an example, the first... The absolute value of the difference between the recording time of the first historical dimension running data and the last historical dimension running data in each running data segment is denoted as the nth segment. The runtime of the first data segment; The running time of the first data segment and the first The ratio of all historical dimensions of running data in the first running data segment is denoted as the i-th segment. The single-segment runtime value of the first running data segment; the first segment's runtime value; The data categories of each dimension are in the first The variance of the single-segment runtime value of all running data segments in the sky, denoted as the i-th segment. The data categories of each dimension are in the first The daily running time value in the sky; all dimensional data types in the first The average daily running time of the day is denoted as the th day. Daily data execution level. Retrieve the daily data execution level for each day.
[0076] Furthermore, the first The data categories of each dimension are in the first Tianzhongdi The mean of the boundary factors of all historical dimensions of running data in each running data segment is denoted as the first running boundary mean. The mean of the boundary factors of the first running data segment is then used. The data categories of each dimension are in the first Tianzhongdi The mean of the boundary factors of all historical dimension running data in each running data segment is denoted as the second running boundary mean. The mean of the first running boundary mean and the mean of the second running boundary mean are denoted as the third running boundary mean. Day The first of the data categories in the dimension The first running data segment and the first The first of the data categories in the dimension The dimension boundary reference value of the first running data segment; the first Day The data types of each dimension and the first The dimensional boundary reference values of all data segments with the same sequence number across all data categories in each dimension are linearly normalized, and each normalized dimensional boundary reference value is recorded as a dimensional boundary reference factor. Based on the daily data execution level, the first data segment within each day... The first of the data categories in the dimension The first running data segment and the first The first of the data categories in the dimension The dimension boundary reference factor for the first running data segment, and the first... The data types of each dimension and the first The correlation between data categories of each dimension and the same sequence number segment among different data categories of each dimension on each day is obtained. The data types of each dimension and the first The degree of dimensional correlation among data types in each dimension. As an example, the degree of correlation can be calculated using the following formula: The data types of each dimension and the first The degree of dimensional correlation among data types in each dimension:
[0077] ;
[0078] In the formula, Indicates the first The data types of each dimension and the first The degree of dimensional correlation among data categories in each dimension; This indicates the total number of days. Indicates the first Daily data execution level; Indicates the first The number of all running data segments in each dimension of the data category in Tianzhong; Indicates the first Day The first of the data categories in the dimension The first running data segment, and the first The first of the data categories in the dimension Dimensional boundary reference factor for each running data segment; Indicates the first Day The first of the data categories in the dimension The first running data segment, and the first The first of the data categories in the dimension Pearson correlation coefficient for each running data segment; This represents an exponential function with the natural constant as its base. The example uses... The model is used to represent the inverse proportional relationship and for normalization processing. As input to the model, implementers can choose the inverse proportional function and the normalization function according to the actual situation; Indicates the first Day Among the data categories of the first dimension, The distribution of data segments across the data categories in each dimension. Where the first... The data types of each dimension and the first The greater the dimensional correlation among the data types in each dimension, the stronger the correlation between the data types in the first dimension. The data types of each dimension and the first The greater the degree to which the data types of each dimension influence each other in the real environment, the more it reflects the... The data types of each dimension and the first The greater the importance of each data category in a dimension for intelligent monitoring, the more important it becomes. The degree of correlation between any two data categories in each dimension is obtained, and all correlations are linearly normalized. Each normalized correlation is then recorded as a dimension correlation factor. Furthermore, obtaining the Pearson correlation coefficient is a well-known technique and will not be elaborated upon in this embodiment.
[0079] Furthermore, any one dimension data type is designated as the target dimension data type, and each dimension data type other than the target dimension data type is designated as a reference dimension data type; taking any one reference dimension data type as an example, a preset dimension correlation factor threshold is defined. In this embodiment, This example is used for illustration; no specific limitations are set in this embodiment. This can be determined based on the specific implementation; if the dimensional correlation factor between the target dimension data type and the reference dimension data type is greater than... The combination of the target dimension data type and the reference dimension data type is denoted as the dimension type combination of the target dimension data type; all dimension type combinations of the target dimension data type are obtained, and all dimension type combinations of each dimension data type are obtained.
[0080] Thus, the above method yields all possible combinations of data types for each dimension.
[0081] Step S004: Based on the similarity of the connection edges between the node structures of each historical dimension running data and the corresponding node structures of other historical dimension running data in the combination of dimension types, monitor the historical dimension running data to obtain several abnormal dimension running data.
[0082] Specifically, any one of the data categories is designated as the first target data category. Taking any combination of data categories within the first target data category as an example, the historical data of each reference data category within that combination is used as the x-coordinate of the node, and the historical data of each historical data category within the first target data category is used as the y-coordinate of the node. Several nodes are constructed based on the x-coordinate and y-coordinate of each node. Taking any two nodes as an example, the inversely proportional normalized value of the Euclidean distance between these two nodes is designated as the node connection determination distance. A preset threshold for the node connection determination distance is established. In this embodiment, This example is used for illustration; no specific limitations are set in this embodiment. It depends on the specific implementation; if the node connection determination distance between the two nodes is greater than... Connect these two nodes and denote the connection as a link; if the node connection distance between the two nodes is less than or equal to... The connection between these two nodes is recorded as a disconnection; all connections between any two nodes are obtained, and the node structure formed by all connections between any two nodes is recorded as the combined node structure of this dimension. For the inverse proportional normalized value of the Euclidean distance between any two nodes, this embodiment uses... The model is used to represent the inverse proportional relationship and for normalization processing. For the model input, the implementer can choose between an inverse proportional function and a normalization function based on the actual situation. It should be noted that each historical dimension of running data corresponds to one node. Furthermore, obtaining Euclidean distance is a well-known technique and will not be elaborated upon in this embodiment.
[0083] Furthermore, in the first target dimension data category, any historical dimension running data is designated as the target running data, and each historical dimension running data other than the target running data is designated as the reference running data. Taking any reference running data and any dimension category of the first target dimension data category as an example, in the combined node structure of this dimension category combination, the node corresponding to the target running data is designated as the target node, the node corresponding to the reference running data is designated as the reference node, each node connected to the target node is designated as the first reference node, and each node connected to the reference node is designated as the first reference node. The node is denoted as the second reference node. Each node that is the same in the first and second reference nodes is denoted as the node on the same side of the target running data and the reference running data. Taking any node on the same side as an example, the vector formed by the target node and the node on the same side is denoted as the first vector, and the vector formed by the reference node and the node on the same side is denoted as the second vector. The cosine similarity between the first vector and the second vector is denoted as the first similarity between the target node and the reference node under the node on the same side. The mean of the first similarity between the target node and the reference node under all nodes on the same side is denoted as the node edge similarity between the target running data and the reference running data.
[0084] Furthermore, based on the node edge similarity between the target running data and the reference running data, the proximity degree between the target running data and the reference running data is obtained. As an example, the proximity degree between the target running data and the reference running data can be calculated using the following formula:
[0085] ;
[0086] In the formula, This indicates the proximity of the target running data to the reference running data; This indicates the number of all dimension category combinations to which the target running data belongs in the first target dimension data category; Indicates the first In a combined node structure of multiple dimension types, the number of all nodes on the same side between the target running data and the reference running data; Indicates the first The number of connections in a combined node structure with multiple dimensions is the total number of connections. Indicates the first In a combined node structure with multiple dimensions, the similarity between the target running data and the reference running data is considered. A higher degree of proximity between the target running data and the reference running data indicates a more similarity in their changes, reflecting a greater similarity in their operation on the intelligent service terminal. The degree of proximity between the target running data and all reference running data is obtained, and all degree of proximity is linearly normalized. Each normalized degree of proximity is recorded as a proximity factor.
[0087] Furthermore, a nearest neighbor factor threshold is preset. In this embodiment, This example is used for illustration; no specific limitations are set in this embodiment. This can be determined based on the specific implementation; the nearest neighbor factor of the target operating data should be greater than [a certain value]. The reference running data is denoted as the nearest neighbor running data. All nearest neighbor running data of the target running data are obtained, as are all nearest neighbor running data of each historical dimension running data. Each historical dimension running data corresponds to multiple nearest neighbor running data.
[0088] Furthermore, based on the Euclidean distance between each nearest neighbor of the historical dimension's running data and each node of the historical dimension's running data, the CABDDCG clustering algorithm based on dynamic splitting of the connected graph is used to cluster all historical dimension's running data, resulting in several final clusters; a preset threshold for the proportion of nodes is also included. In this embodiment, This example is used for illustration; no specific limitations are set in this embodiment. This can be determined based on the specific implementation; the number of all historical dimension operation data is recorded as... ,Will and The product of these is denoted as the quantity threshold. Taking any final cluster as an example, if the number of historical data points in all dimensions of the final cluster is less than... Each historical dimension of the running data in the final cluster is recorded as the abnormal dimension running data; all abnormal dimension running data are then obtained. The process of clustering based on the Euclidean distance between nodes is a well-known aspect of the CABDDCG clustering algorithm based on dynamic splitting of connected graphs, and will not be elaborated upon in this embodiment. Please refer to... Figure 2 It shows a flowchart of the characteristic relationships of an intelligent monitoring method for the operation of intelligent service terminals.
[0089] This concludes the embodiment.
[0090] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for intelligently monitoring the operation of an intelligent service terminal, characterized in that: The method comprises the following steps: Collecting a number of historical dimension operation data of each dimension data type in a number of days, wherein the historical dimension operation data corresponds to a recording time; According to the operation cycle changes and operation trend differences of the historical dimension operation data of the same dimension data type on the same day, the demarcation factor of each historical dimension operation data is obtained; according to the demarcation factor, several historical dimension operation data of each dimension data type in several days are divided to obtain several operation data segments; According to the association between different dimensional data types of operation data segments with the same sequence number, the change of data operation rules, the demarcation factor of each corresponding historical dimensional operation data and the recording time, several dimensional data types are combined to obtain several dimensional type combinations of each dimensional data type; According to the similarity of the connection edges between each historical dimension operation data and the corresponding node structures of other historical dimension operation data in the node structure of the several historical dimension operation data in the dimension type combination, the several historical dimension operation data are monitored to obtain several abnormal dimension operation data; The method of combining several dimensional data types according to the association between different dimensional data types of operation data segments with the same sequence number, the change of data operation rules, the demarcation factor of each corresponding historical dimensional operation data and the recording time to obtain several dimensional type combinations of each dimensional data type includes the following specific methods: For the f-th operation data segment of the a-th dimension data type on the b-th day, the absolute value of the difference in recording time between the first historical dimension operation data and the last historical dimension operation data in the f-th operation data segment is recorded as the operation time of the f-th operation data segment; the ratio of the operation time of the f-th operation data segment to all historical dimension operation data in the f-th operation data segment is recorded as the single-segment operation time value of the f-th operation data segment; the variance of the single-segment operation time values of all operation data segments of the a-th dimension data type on the b-th day is recorded as the single-day operation time value of the a-th dimension data type on the b-th day; the mean of the single-day operation time values of all dimension data types on the b-th day is recorded as the single-day data operation degree on the b-th day; Record the mean of the demarcation factors of all historical dimensional operating data in the f-th operating data segment of the a1-th dimensional data type on the b-th day as the first operating demarcation mean, record the mean of the demarcation factors of all historical dimensional operating data in the f-th operating data segment of the a2-th dimensional data type on the b-th day as the second operating demarcation mean, and record the mean of the first operating demarcation mean and the second operating demarcation mean as the dimensional demarcation reference value of the f-th operating data segment in the a1-th dimensional data type and the f-th operating data segment in the a2-th dimensional data type on the b-th day; linearly normalize the dimensional demarcation reference values of all operating data segments with the same sequence number between the a1-th dimensional data type and the a2-th dimensional data type on the b-th day, and record each dimensional demarcation reference value after normalization as the dimensional demarcation reference factor; According to the daily data operation degree of each day, the dimension boundary reference factor of the f-th operation data segment in the a1-th dimension data type and the f-th operation data segment in the a2-th dimension data type on each day, and the association of the same sequence number segments between the a1-th dimension data type and the a2-th dimension data type in different dimension data types on each day, the dimension association factor between the a1-th dimension data type and the a2-th dimension data type is obtained; Any dimensional data type is recorded as a target dimensional data type, and each dimensional data type other than the target dimensional data type is recorded as a reference dimensional data type; for any reference dimensional data type, a dimension correlation factor threshold value T5 is preset, and if the dimension correlation factor between the target dimensional data type and the reference dimensional data type is greater than T5, the combination of the target dimensional data type and the reference dimensional data type is recorded as the dimensional type combination of the target dimensional data type; The dimension association factors are specifically: The dimensional boundary reference factor between the f-th running data segment in the a1-th dimensional data type on the b-th day and the f-th running data segment in the a2-th dimensional data type is recorded as the first reference factor; the Pearson correlation coefficient between the f-th running data segment in the a1-th dimensional data type on the b-th day and the f-th running data segment in the a2-th dimensional data type is recorded as the first correlation coefficient between the a1-th dimensional data type and the a2-th dimensional data type on the b-th day; the mean of the first correlation coefficients between the a1-th dimensional data type and the a2-th dimensional data type on all days is recorded as the fourth mean; the product of the inversely proportional normalized value of the single-day data running degree on the b-th day and the fourth mean is recorded as the second product of the a1-th dimensional data type and the a1-th dimensional data type on the b-th day; the mean of the second product of the a1-th dimensional data type and the a1-th dimensional data type on all days is recorded as the dimensional correlation degree between the a1-th dimensional data type and the a2-th dimensional data type; The dimensional correlation degree between any two dimensional data types is obtained, all dimensional correlation degrees are linearly normalized, and each dimensional correlation degree after normalization is recorded as the dimensional correlation factor.
2. The method for intelligent monitoring of the operation status of an intelligent service terminal according to claim 1, characterized in that: The method of obtaining the demarcation factor of each historical dimension operation data according to the operation cycle change and operation trend difference of the historical dimension operation data of the same dimension data type on the same day includes the following specific methods: Preset an initial window size T1, a step size T2, and a termination window size T3; take the initial window size T1 as the starting point of the window size, with a step size of T2, and sequentially expand the window size until the expanded window size meets T3, then stop expanding the window, obtain the window size in each expansion process, and record the window size in each expansion process as the time decomposition window size; For any time decomposition window size, obtain several clusters of the a-th dimension data type under the z-th decomposition item under the time decomposition window size; obtain similar clusters of each cluster of the a-th dimension data type under the z-th decomposition item under the time decomposition window size; In the cluster to which the zth decomposition item of the ath dimensional data type on the bth day belongs, the set consisting of all decomposition items except the zth decomposition item of the ath dimensional data type on the bth day is recorded as the reference decomposition set of the zth decomposition item of the ath dimensional data type on the bth day; according to the historical dimensional operation data contained in different clustering clusters under different decomposition items of the ath dimensional data type on the bth day, the reference decomposition set and the corresponding similar clustering clusters, the decomposition window weight of the ath dimensional data type on the bth day under the time decomposition window size is obtained; For the kth historical dimensional operating data of the ath dimensional data type on the bth day, obtain the boundary degree of the kth historical dimensional operating data of the ath dimensional data type on the bth day according to the numerical change of all historical dimensional operating data of the ath dimensional data type on the bth day and the decomposition window weight of the ath dimensional data type on the bth day under each time decomposition window size; obtain the boundary degree of each historical dimensional operating data of all dimensional data types on each day, linearly normalize all the boundary degrees, and record each normalized boundary degree as a boundary factor.
3. The method for intelligent monitoring of the operation status of an intelligent service terminal according to claim 2, characterized in that: The specific method of obtaining a number of clusters of the ath dimension data type under the zth decomposition item under the time decomposition window size includes: The time decomposition window size is used as the window size. According to the window size, the STL time series decomposition algorithm is used to decompose the sequence composed of all historical dimension operation data of the a-th dimension data type on the b-th day, and the trend item, cycle item and residual item of the a-th dimension data type on the b-th day are obtained; and the trend item and cycle item of the a-th dimension data type on the b-th day are recorded as the decomposition items of the a-th dimension data type on the b-th day respectively; By using the least squares method, curve fitting is performed on all historical dimensional operating data in the zth decomposition item of the ath dimensional data type on the bth day to obtain the fitting curve of the zth decomposition item of the ath dimensional data type on the bth day; the fitting curve of the zth decomposition item of the ath dimensional data type on all days is obtained, and the inverse proportional value of the Fréchet distance of the fitting curve between the zth decomposition items of the ath dimensional data type on any two days is recorded as the decomposition identification distance, and the decomposition identification distance is used as the distance metric. According to the distance metric, the DBSCAN clustering algorithm is used to cluster the zth decomposition items of the ath dimensional data type on all days to obtain several clustering clusters under the zth decomposition item of the ath dimensional data type.
4. The method for intelligent monitoring of the operation of an intelligent service terminal according to claim 2 or 3, characterized in that: The specific method of obtaining similar clusters of each cluster of the ath dimension data type under the zth decomposition item under the time decomposition window size includes: Any cluster of the a-th dimension data type under the z-th decomposition item is recorded as a marked cluster, and each cluster except the marked cluster is recorded as a reference cluster; for any reference cluster, the cluster combination formed by the marked cluster and the reference cluster is recorded as the first cluster combination of the marked cluster, and the average value of the Fréchet distance of the fitting curve between all any two decomposition items between the marked cluster and the reference cluster is recorded as the similarity coefficient of the first cluster combination; The similarity coefficients of all first clustering cluster combinations are obtained, and the first clustering cluster combination with the smallest similarity coefficient is recorded as a similar first clustering cluster combination; and the reference cluster in the similar first clustering cluster combination is recorded as a similar cluster of the target cluster.
5. The method for intelligent monitoring of the operation of an intelligent service terminal according to claim 2 or 3, characterized in that: The method of obtaining the decomposition window weight of the ath dimensional data type on the bth day under the time decomposition window size according to the historical dimensional operation data contained in different clustering clusters under different decomposition items of the ath dimensional data type on the bth day, the reference decomposition set and the corresponding similar clustering clusters includes: The mean of the decomposition identification distances between all decomposition items in the reference decomposition set of the zth decomposition item of the ath dimensional data type on the bth day and the zth decomposition item of the ath dimensional data type on the bth day is recorded as the first mean; the mean of the decomposition identification distances between all decomposition items in the similar clustering cluster to which the zth decomposition item of the ath dimensional data type on the bth day belongs and the zth decomposition item of the ath dimensional data type on the bth day is recorded as the second mean; the ratio of the first mean to the second mean is recorded as the first ratio of the zth decomposition item of the ath dimensional data type on the bth day; the mean of the first ratios of all decomposition items of the ath dimensional data type on the bth day is recorded as the decomposition window weight of the ath dimensional data type on the bth day under the time decomposition window size.
6. The method for intelligent monitoring of the operation of an intelligent service terminal according to claim 2 or 3, characterized in that: The method of obtaining the demarcation degree of the kth historical dimensional operation data of the ath dimensional data type on the bth day according to the numerical changes of all historical dimensional operation data of the ath dimensional data type on the bth day and the decomposition window weight of the ath dimensional data type on the bth day under each time decomposition window size includes the following specific methods: For any two adjacent historical dimension running data of the a-th dimension data type on the b-th day, in these two historical dimension running data, the ratio of the difference between the second historical dimension running data and the first historical dimension running data and 2 is recorded as the slope of the second historical dimension running data; obtain the slope of all historical dimension running data of the a-th dimension data type on the b-th day; The autocorrelation function is used to obtain the period of the periodic items of the a-th dimension data type in each day under the time decomposition window size, and the absolute value of the difference between the periodic items of the a-th dimension data type in any two days is used as the distance metric. According to the distance metric, the DBSCAN clustering algorithm is used to cluster the periodic items of the a-th dimension data type in each day under the time decomposition window size to obtain several clusters, and each cluster is recorded as a dimension period cluster cluster; The demarcation degree of the kth historical dimension operation data of the ath dimension data type on the bth day under the qth time decomposition window size is recorded as the first demarcation degree; the demarcation degree of the vth historical dimension operation data other than the kth historical dimension operation data within the dimension period clustering cluster to which the kth historical dimension operation data belongs on the bth day under the qth time decomposition window size is recorded as the second demarcation degree; the absolute value of the difference between the first demarcation degree and the second demarcation degree is recorded as the first absolute value of the vth historical dimension operation data; the mean of the first absolute values of all historical dimension operation data other than the kth historical dimension operation data within the dimension period clustering cluster to which the kth historical dimension operation data belongs on the nth day under the qth time decomposition window size of the ath dimension data type is recorded as the third mean; the maximum value of the demarcation degree of all historical dimension operation data in all dimension period clustering clusters to which the ath dimension data type belongs under the qth time decomposition window size is recorded as the first maximum value; the ratio of the third mean to the first maximum value is recorded as the second ratio; The sum of the residual term of the a-th dimensional data type on the b-th day under the q-th time decomposition window size and 1 is recorded as the first sum value; the decomposition window weight of the a-th dimensional data type on the b-th day under the q-th time decomposition window size is recorded as the first weight; the product of the variance of the slope of all historical dimensional operating data of the a-th dimensional data type on the b-th day, the first weight and the first sum value is recorded as the first product of the k-th historical dimensional operating data of the a-th dimensional data type on the b-th day under the q-th time decomposition window size; the mean of the first product of the k-th historical dimensional operating data of the a-th dimensional data type on the b-th day under all time decomposition window sizes is recorded as the boundary degree of the k-th historical dimensional operating data of the a-th dimensional data type on the b-th day.
7. The method for intelligent monitoring of the operation status of an intelligent service terminal according to claim 1, characterized in that: The method of dividing the historical dimensional operation data of each dimensional data type in a number of days according to the demarcation factor to obtain a number of operation data segments includes: A demarcation factor threshold T4 is preset; the historical dimension operation data with a demarcation factor greater than T4 is recorded as the initial segmentation point; the recording time corresponding to each initial segmentation point is recorded as the segmentation time point, and the data segment composed of the historical dimension operation data between adjacent segmentation time points of each dimension data type on each day is recorded as the operation data segment of each dimension data type on each day.
8. The method for intelligent monitoring of the operation status of an intelligent service terminal according to claim 1, characterized in that: The method of monitoring the historical dimension operation data according to the similarity of the connection edges between each historical dimension operation data and the corresponding node structures of other historical dimension operation data in the node structure to obtain the abnormal dimension operation data includes the following specific methods: Record any one dimension data type as the first target dimension data type. For any one dimension type combination of the first target dimension data type, use each historical dimension operation data in the reference dimension data type of the first target dimension data type in the dimension type combination as the horizontal coordinate of the node, and use each historical dimension operation data in the first target dimension data type as the vertical coordinate of the node. Construct several nodes according to the horizontal coordinate and the vertical coordinate of the node. For any two nodes, record the inversely proportional normalized value of the Euclidean distance between the two nodes as the node connection determination distance. A node connection determination distance threshold T6 is preset. If the node connection determination distance of the two nodes is greater than T6, the two nodes are connected, and the connection relationship of the two nodes is recorded as connected; if the node connection determination distance of the two nodes is less than or equal to T6, the connection relationship of the two nodes is recorded as disconnected; the connection relationship between all arbitrary two nodes is obtained, and the node structure formed by the connection relationship between all arbitrary two nodes is recorded as a combined node structure of the dimension type combination; In the first target dimension data type, any historical dimension operation data is recorded as the target operation data, and each historical dimension operation data except the target operation data is recorded as the reference operation data; for any reference operation data and any dimension type combination of the first target dimension data type, in the combined node structure of the dimension type combination, the node corresponding to the target operation data is recorded as the target node, the node corresponding to the reference operation data is recorded as the reference node, each node that is connected to the target node is recorded as the first comparison node, each node that is connected to the reference node is recorded as the second comparison node, and each identical node in the first comparison node and the second comparison node is recorded as the same-side node of the target operation data and the reference operation data; for any same-side node, the vector formed by the target node and the same-side node is recorded as the first vector, the vector formed by the reference node and the same-side node is recorded as the second vector, and the cosine similarity between the first vector and the second vector is recorded as the first similarity between the target node and the reference node under the same-side node; the average of the first similarities between the target node and the reference node under all same-side nodes is recorded as the node edge similarity between the target operation data and the reference operation data; The number of all nodes on the same edge of the target running data and the reference running data in the combined node structure of the w-th dimension type combination is recorded as a first number; The number of all connection relationships in the combined node structure of the w-th dimensional type combination is recorded as the second number; the ratio of the first number to the second number is recorded as the third ratio; the node similarity between the target operating data and the reference operating data in the combined node structure of the w-th dimensional type combination is recorded as the reference similarity; the product of the third ratio and the reference similarity is recorded as the third product of the w-th dimensional type combination of the first target dimensional data type to which the target operating data belongs; the average of the third products of all dimensional type combinations of the first target dimensional data type to which the target operating data belongs is recorded as the degree of proximity between the target operating data and the reference operating data; Obtain the neighboring degrees of the target running data and all the reference running data, linearly normalize all the neighboring degrees, and record each normalized neighboring degree as a neighboring factor; A neighbor factor threshold T7 is preset, and reference operation data whose neighbor factor with the target operation data is greater than T7 is recorded as neighbor operation data, and all neighbor operation data of the target operation data are obtained, and all neighbor operation data of the operation data of each historical dimension are obtained; According to the Euclidean distance between each neighboring operation data of each historical dimension operation data and each historical dimension operation data for the node, all historical dimension operation data are clustered using the CABDDCG clustering algorithm based on dynamic splitting of the connected graph to obtain several final clustering clusters; a node quantity ratio threshold T8 is preset, the number of all historical dimension operation data is recorded as J, and the product of J and T8 is recorded as the quantity threshold T9; for any final clustering cluster, if the number of all historical dimension operation data in the final clustering cluster is less than T9, each historical dimension operation data in the final clustering cluster is recorded as abnormal dimension operation data.
